A matched case-control study in Taiwan to evaluate potential risk factors for prostate cancer

The rising incidence rate of prostate cancer (PCa) worldwide has become a public health concern. PCa has a multifactorial etiology, and the link between human papillomavirus (HPV) and PCa has been widely investigated by numerous case–control studies. This age-matched, case–control study included 143 PCa patients and 135 benign prostatic hyperplasia (BPH) patients, with prostatic specimens testing negative for malignancy, as control. Study participants were recruited from four major hospitals in Taoyuan City, Taiwan, period 2018–2020, looking into HPV infection and other PCa risk factors, including dietary habits, family history, personal lifestyle, and sexual behavior. Multiple logistic regression analysis and forward stepwise selection analysis were conducted to identify potential risk factors for PCa. HPV DNA was found in 10 of the 143 PCa cases (7%) and 2 of the 135 BPH controls (1.5%) (OR = 6.02, 95% CI = 1.03–30.3, p = 0.046). This association was slightly significant, and furthermore, high risk HPV was not found to be associated with PCa. Higher body mass index (BMI) (OR = 1.15, 95% CI = 1.05–1.27, p = 0.003), more total meat consumption (OR = 2.74, 95% CI = 1.26–5.94, p = 0.011), exhibited association to PCa. However, PCa family history only presented a statistically significant difference by forward stepwise analysis (OR = 3.91, 95% CI = 1.17–13.12, p = 0.027). While much focus has been on the association between HPV and PCa, the results of this study indicate that more efforts should be directed towards investigating dietary habits, personal lifestyle and family history as factors for PCa. These results could serve as a basis for designing PCa prevention strategies.


Methods
Study design. This study utilized a prospective hospital-based, matched case-control design, at a 1:1 ratio.
The case group consisted of 143 PCa patients, while the control group consisted of 135 BPH patients with prostatic biopsies testing negative for malignancy. Study participants were recruited between February 2018 and December 2020, from four major hospitals in Taoyuan City, Taiwan, upon approval from with each's respective institutional review board (IRB). Patient age at enrollment was between 55 and 86. According to Taiwan's Cancer Registry Annual Report for 2017, the median diagnostic age of PCa in Taiwan was 73 7 , whilst concerning incidence rates of PCa and BPH was relatively low for ages 55 years and younger. Patients aged 86 and over, were excluded due to perceived slow reaction times, poor physical performance statuses, and difficulties in comprehension of interview problems. Age was matched ± 5 years range between case and control groups. Exclusion criteria (assessed via questionnaire interview) included HPV-vaccinated as well as patients not initially diagnosed as PCa.

Measure of exposures.
We collected laboratory data from patients who agreed to provide prostate specimens at the urology departments of any of the four hospital sites. The main surgical procedures were radical prostatectomy (RP), transurethral resection of the prostate (TURP), and transrectal ultrasound biopsy (TRUS biopsy). RP was only performed in the PCa group. Informed consent forms were required. Potential PCa risk factors analyzed include detection of HPV infection, as well as dietary habits, family history, personal lifestyle, and sexual behavior. A structuralized questionnaire was devised to identify for possible covariates, such as risk factors and exposures, during which 15 questionnaires and two experts were consulted. Based on an IRB request for patient privacy, the option to refuse answering for individual questions was allowed. The collected information included age, body mass index (BMI), diet, family history of cancer, occupational record, race, history of sexual activity and STDs, as well as use of tobacco, alcohol, and betelnut chew. The questionnaire was administered by qualified interviewers in-person. Only records of patients who provided both pathological specimens and completed the questionnaire were included in the data analysis.

Measure of outcomes.
Cases ascertained were those pathologically proven for PCa, aged 55 to 86. The majority of these cases had biopsies due to elevated-PSA and were suspected of PCa. A small number of cases were incidentally found to be malignant PCa during TURP for BPH symptoms. Selection of controls, or confirmed non-cases, were defined as BPH patients with clinically apparent lower urinary tract symptoms (LUTS), and were refractory to oral medicines, urethral catheterization, and received TURP from urologists. Another source of non-cases were patients with incidental PSA-elevation and highly suspected of PCa, whose TRUS biopsies detected no cancer cells.
The FFPE tissue specimen blocks were collected from the participating hospitals, deriving from different types of surgical procedures including RP, TRUS biopsy, and TURP. These FFPE blocks were sent to the Taipei Institute of Pathology where H&E staining slides had been prepared for pathologist review to confirm the diagnosis of malignant or benign disease. A minimum of three tissue sections, 5 um in thickness, per patient were collected for DNA extraction. DNA was extracted by using MagCore ® Genomic DNA Tissue kit (Cartridge Code 401, RBC Bioscience).

Statistical analysis.
Control and case groups were matched for age, and chi-square test was utilized to verify that there was no difference in distribution by age groups. Twenty patients from each group were selected for a second interview, within 30 days of each patient's first interview, to validate the reliability of the questionnaire. Test-retest reliability coefficient was computed to measure the correlation between results from the first and second interview. Intraclass correlation coefficients (ICC) value of > 0.7 was deemed to be acceptable. As for the exposures or potential risk factors for PCa, independent t-test and chi-square test were adopted for univariate analysis of all variables and PCa status. If any individual variables reached statistical significance (p < 0.05), further multiple logistic regression analysis was conducted to determine the effect size of such single independent variables, in the presence of actual or potential confounding. Forward stepwise selection is a variable selection method, starting with a model that contains no variables (null model), then gradually adding the most significant variables one at a time, until it either reaches the stopping rule (criterion) or includes all of the variables into the model. The stopping rule (criterion) in this study is when p ≥ 0.05. The most significant variables are chosen from the result of simple logistic regression, including BMI, HPV detection, occupation exposure, etc.
Reliability. The consistency of responses to the questionnaire was examined as follows: 20 cases and 20 controls, interviewed twice within 30 days, conducted by the same interviewer. Test-retest reliability was estimated by calculating the (CC) of the measured values at two separate time points. A higher CC between measured data denotes greater test-retest reliability, whereas Pearson's r-value ≧ 0.7 indicates acceptable correlation. Uniformity of questionnaire interviews was enhanced by selecting one Principal Investigator (PI) and one assistant per hospital site, whereby the assistant was trained by the PI to conduct interview in similar manner.
Ethical approval and consent to participate. This study was performed in line with the principles of the Declaration of Helsinki. We obtained IRB approval from National Yang Ming Chiao Tung University, Min Sheng General Hospital, Landseed International Hospital, Taoyuan General Hospital and Ten Chen Hospital. Informed consent was obtained from all individual participants included in the study. The chi-square test showed that the age distribution between groups were not different, with p value of 0.826, substantiating the age-matching method. Mean subject height was 164.8 cm (± 5.9 SD) for case group, and 166.7 cm (± 6.0 SD) for control group, with significant difference (p = 0.007). Mean subject weight was 68.2 kg (± 9.4 SD) in case group and 66.2 kg (± 10.1 SD) in control group, with no significant difference (p = 0.091). The mean BMI was 25.7 kg/m 2 (± 3.2 SD) in cases and 23.8 kg/m 2 (± 3.3 SD) in controls, with significant difference (p < 0.001). In terms of occupational exposure, i.e., chemical solutions/materials, plastic materials, metal dust, machinery, electroplating, printing ink, dyes, paints, radiation, and electromagnetic field, subjects in PCa group had a significantly higher proportion than patients in BPH group (33.8% vs 19.3%, p = 0.006). BPH was more www.nature.com/scientificreports/ remarkable in control group than in case group, 94.8% versus 79.0% (p < 0.001), as selection for control group was largely from BPH patients with clinically apparent LUTS, as well as patients with incidentally determined PSA elevation. We detected 10 out of 143 PCa patients as well as 2 out of 135 BPH patients to be HPV positive, with positive detection rate of 7% versus 1.5% (p = 0.024), which demonstrates greater HPV correlation in the PCa group. Revealed HPV types were 18 (2), 52 (1), 53 (3), 62 (1), non-16/18 (3) in case group, while type 52 (2) was only noted in control group, with number in parenthesis denoting number of HPV cases. HPV types 18, 52, 58, 62 were high risk while non-16/18 was low risk. However, the association between total HPV detection and PCa was slightly significant in further logistic regression analysis. If low risk types were not included, then high risk HPV detection was not associated with PCa.
Other non-significant variables including education level of subjects, marital status, working status, as well as paternal and maternal tongue were shown in supplements (Table S1).
Medical history. History of radiation therapy (RT), dietary supplement, and medication history was illustrated in Table S2. RT history was defined as any radiation treatment for any cancer type other than PCa prior to recruitment, where no difference was found between the groups. Popular dietary supplements among adults in Taiwan include calcium, vitamin B, vitamin C, vitamin complex, zinc, selenium, green tea powder, isoflavone, royal jelly, phytoestrogen essential, iron, omega-3 fatty acid, vitamin D, glucosamine, probiotic, lutein, chicken essence, amongst others. We observed no difference in dietary supplement intake between two groups. Numerous patients had history of taking medications, and there was no difference observed in drug history. Chronic inflammation was proved to induce carcinogenesis in many ways, and PCa was also hypothesized to be correlated with pelvic surgery, pelvic disease, or other status related with chronic inflammation. However, we observed no significant difference in urinary calculi (p = 0.616), urinary tract infection (p = 0.174), bladder disease (p = 0.833), pelvic trauma (p = 0.859), pelvic surgery (p = 0.709), autoimmune disease (p = 0.486), pelvic cancer (p = 0.531), and extrapelvic cancer (p = 0.061) between the groups (Table S3).
Personal lifestyles. Differences in personal lifestyles, especially in dietary habit, were noted (Table 1). Meat consumption ≧ 1 serving per meal was 81.8% in the case group and 49.6% in the control group, (p < 0.001). Meat consumption was categorized as white meat, seafood, and red meat. The majority of subjects in both groups ate red meat primarily, but red meat consumption was significantly higher in case group than control group (83.35% vs 69.3%, p = 0.008). We further explored meat consumption by questioning subjects as to percentage of total meat consumed being red meat, as stratified into three groups: less than 25%, between 25 and 50%, and greater than 50%. Case group was 22.7%, 31.8%, and 45.5%, respectively, whereas control group was 45.7%, 32.3%, and 22.0% (p < 0.001). Case group also reported greater update of dairy products, although not statistically significant (p = 0.052). Case group also consumed fewer vegetables (p = 0.01), fewer fruits (p = 0.027), and fewer soybean products than the control group. An interesting observation was that all vegetarian subjects belonged to the control group, while none were found in the case group (p < 0.003), in line with the possibility that vegetarian diets are a preventative factor for PCa.
Other non-significant variables were shown in Table S4. Consumption of tea drink, alcohol and tobacco were similar in both groups. Endemic to Taiwan and other parts of East Asia is betelnut chewing, which has strong association with oral cancer and other systemic effects. However, we did not observe the correlation between betelnut habit and PCa in this study (p = 0.173).
Reproductive history and family history. Family history of PCa is a complex mix of genetic and environmental factors, and it is an important parameter that can be assessed in clinical practice through basic questionnaires 30 . This study also explored family history of PCa via questionnaire, and results revealed positive correlation between PCa and family history of PCa, with 11.4% in case group and 3.7% in control group, with family history of PCa (p = 0.016) ( Table 2). However, further investigation (Table S5) into which family member had diagnosis of PCa, such as grandfather, granduncle, father, uncle, brother, cousin, son, or grandson, did not show significance perhaps due to inadequate sample size. International Prostate Symptom Score (IPSS) is a tool with 7 questions related to different urinary symptoms (incomplete emptying, frequency, intermittency, urgency, weak stream, and straining), which allows urologists to better understand the severity of BPH. IPSS can score from 0 to 35. In Table 1, the mean IPSS was 14.1 (± 7.8 SD) in case group and 17.3 (± 7.3 SD) in control group with difference (p < 0.001), and score summation 8-19 was deemed as moderate BPH. Circumcision before first sexual intercourse was associated with a slightly lower reduction in the relative risk of PCa in some studies 31 , but inconclusive. Our study showed contradictory results, with more patients in case group than control having received circumcision (20.0% vs 9.6%, p = 0.016) ( Table 1). However, more than 90% of patients in our study had circumcision after adulthood, but not before first sexual intercourse.
The two groups were not different in regard to history of androgen replacement therapy (for hypogonadism) or anti-alopecia agent (such as finasteride) (Table S6). Vasectomy was linked to a slight increase in long-term risk of PCa in some studies 32 , but this issue remains controversial. In our study, 3.6% of case group and 7.4% of control group have received vasectomies, without significant difference (p = 0.161). The condition of genital beading and tattoo over lower abdomen or genital organ, and the experience of HPV vaccination was not different between groups. International Index of Erectile Function (IIEF-5), consisting of 5 questions, has evolved to be the standard for evaluating the severity of erectile dysfunction (ED). The 5 domains are erectile function, orgasmic function, sexual desire, intercourse satisfaction, and overall satisfaction, with score ranging from 5 to 25. IIEF-5 was 8.1 (± 9.5 SD) in PCa group and 8.3 (± 9.5 SD) in BPH group, without difference (p = 0.842), and score summation 8-11 was defined as moderate ED. We also questioned subjects regarding sexual behavior and reproductive status of their wives (Tables 2 and S88). We found that the case group had more spouses in lifetime (p = 0.029) and sexual partners (p = 0.037). History of STD in spouses (p = 1.000) and cancer history (p = 0.904) was not different between two groups. HPV vaccines were launched in Taiwan circa 2006, and generally applied to adolescents and young adults, so almost none of the wives in either group had received vaccination for HPV (p = 0.235).

Logistic regression and other analysis.
Results of simple and multiple logistic regression (LR) of risk factors associated with PCa were shown (Table 3). PCa risk was found to be significantly associated with higher BMI (p < 0.001 and p = 0.003), HPV positivity (p = 0.040 and p = 0.046), and meat consumption (p < 0.001 and p = 0.011) in both simple and multiple LR. IPSS summation score of PCa was lower than that of BPH, which was due to essence and definition of controls selection. PCa-associated risk factors, in only simple LR but not in multiple LR, were occupational exposure (p < 0.001), eating red meat as main protein (p = 0.009), higher red meat consumption (p < 0.001), PCa family history (p = 0.022), circumcision history (p = 0.018), ≧ 4 lifetime sexual partners (p = 0.002), suffering 1 type of STD (p = 0.002), wife's lifetime spouse ≧ 2 (p = 0.036), and wife's lifetime sexual partners ≧ 2 (p = 0.044). Potential protective factors for PCa, in only simple LR but not in multiple LR, were daily consumption of more vegetables (p = 0.006), fruits (p = 0.011), and soybean products (p = 0.002). Further forward stepwise LR (Table 4)  Test-retest reliability of the questionnaire was assessed by calculating the correlation coefficient (CC), and all the information was collected in supplement (Table S9). Twenty cases, 20 controls, and a sum of 40 patients were evaluated. Among the 55 evaluable questions, all Pearson correlation coefficients (r) in 40 patients were ≧ 0.7. In the case group, (r) for all 20 subjects were ≧ 0.7, while in the control group, 96.4% (53/55) were ≧ 0.7. Overall result proved acceptable correlation.

Discussion
To our knowledge, this is the first age-matched case-control study in East Asia to look at HPV infection, dietary habit, and other potential risk factors for PCa, using pathology specimens with questionnaires, sample size calculation, logistic regression analysis, and test-retest method.
Clinical studies have suggested that certain eating habits and dietary supplements may prevent or promote PCa. Our research did not find correlation between any specific dietary supplement and PCa, but multiple logistic regression analysis uncovered that BMI and total meat consumption was correlated with PCa. A growing body of literature has linked high intake of meat, particularly red meat and processed meat, to an increased risk of cancer. Evidence showed that this danger was not caused by meat per se, but rather by high-fat consumption or www.nature.com/scientificreports/ mutagenic heterocyclic amines produced by the ways of high-temperature cooking (grilled, barbequed, or deep frying) and processing procedures (smoking, seasoning or curing) 9,33 . Our data showed total meat consumption ≧ 1 serving per meal was at higher PCa risk than those < 1 serving per meal (OR = 2.74, 95% CI: 1.26-5.94, p = 0.011), but interestingly, the risk was not observed for red meat consumption. Methods of meat preparation were not investigated in this study. People who prefer high calorie intake of saturated animal meat and fat have often been associated with an increased risk of obesity and PCa, as subsequent increased testosterone levels are involved in prostate tissue proliferation 9 . Obesity may not only increase PCa incidence, but also increase risk of PCa mortality and recurrence 9,11,28 . Another mechanism is that obesity, especially when combined with physical inactivity, causes a decreased tissue response to insulin, notably in terms of glucose absorption. Insulin resistance causes chronically high blood insulin levels, which is a growth-promoting hormone and hence a scientifically reasonable risk factor for PCa. Obesity, deemed to have more metabolically active adipose tissue, is usually measured by higher BMI or larger waist circumference 11 . The BMI of subjects were also assessed in our study, noting higher BMI yielded slightly higher risk of PCa (OR = 1.15, 95% CI: 1.05-1.27, p = 0.003). Family history of PCa influences the risk of developing the disease 14 . Five single-nucleotide polymorphisms have been identified as having a statistically significant association with PCa when present in conjunction with a family history of PCa, according to Zheng et al. 12 . From 1986 to 2004, the Health Professional Follow-Up Study followed 3695 PCa patients and discovered increased PCa risk by 2.3 times, with a family history of PCa in both the father and a brother (95% CI: 1.76-3.12) 9 . Further evidence from twin studies elucidated that shared genetic factors were responsible for a large portion of familial aggregation of PCa, with a heritability estimate of 75% 11 . The data obtained from the forward stepwise logistic regression method in our study also revealed that those with a PCa family history had a higher risk of PCa (OR = 3.91, 95% CI: 1.17-13.12, p = 0.027). Sometimes, collinearity happens when different independent variables we use in a regression model are highly correlated with one another. Collinearity will inflate the variance and standard error of coefficient estimates. In other words, these independent variables with high correlation explain some of the same variance in the dependent variable, which reduces statistical significance. As a result, the model will become less reliable. One of the solutions to reduce or remove collinearity is to carefully select the variables or consider using other regression methods such as stepwise selection analysis. Collinearity may explain why p-value of PCa family history (0.027 and 0.074) was lower in forward stepwise selection analysis than in multiple logistic regression model. All vegetarians were found in control group (8/135), while none were in PCa group (0/143), indicating that vegetarian lifestyle might have a protective effect for PCa (p = 0.003). Tantamango-Bartley et al. 34 reported that vegan diets may confer a lower risk of PCa (HR = 0.65, 95% CI: 0.49-0.85), since vegans consumed a diet heavy of antioxidant-rich foods, reversing chronic inflammation induced by oxidative stress and free radical production that are known to damage genomic DNA and hasten PCa development. Another study from Taiwan conducted by Chen et al. 35 25 detected HPV DNA in 41.5% (17/41) carcinoma samples, whereas all 30 hyperplasia samples were HPV-negative, but most HPV types were low risk or undefined. Serth et al. 26 detected significantly higher copy numbers of HPV16-E6 sequences in the prostate tumors with 21% (10/47) when compared to the control tissue with 3% (1/37).
In our study, HPV DNA PCR amplification tested positive in 12 of total 278 patients (4.3%), of which 10 were amongst the 143 PCa patients (7%) and 2 amongst the 135 BPH patients (1.5%), with significant difference (OR = 6.02, 95% CI = 1.03-30.3, p = 0.046), suggesting association between HPV and PCa. However, CI of 95% , showed great discrepancy to that observed on our study (7%). Discrepancies in distribution and prevalence of HPV genotypes vary greatly across ethnic groups, geographic areas, and lifestyle groups, may also factor here. This is an observed limitation to our study, and future studies should re-calculate sample size for more robust results. Sexual activity and STDs have been hypothesized to play a role in PCa development via a variety of pathophysiological routes, but the studies have thus far reported mixed results [13][14][15][16][17] . The multiple logistic regression in our study did not reveal significant difference in sexual behavior and STD history. Only multiple and lasting episodes of STDs could cause recurrent inflammation like chronic prostatitis and lead to escalating risk of PCa 37 . Although, in our study, 19.7% (28/142) of PCa patients had STD experience, which was higher than the 8.2% (11/135) in control group (p = 0.003), only a few had multiple lifelong STD episodes (1.4% amongst cases, and 3.0% amongst controls). Furthermore, almost all STD patients had sought prior medical treatment and achieved remission of the infection. These reasons may explain the insignificant results in the multivariate analysis.
Calculation of sample size, test-retest method, matching and adjustment for confounders are common techniques and an important part of study design in epidemiologic research. To our knowledge, these methods were not commonly used in prior case-control studies investigating correlation between HPV and PCa. Complete study design with these methods could be considered as strength of this study. Test-retest method is used to ensure the stability and reliability of survey scores across two time-points over a short period. If reliability is found lacking, there is little confidence that the data produced is an actual representation of the respondents. Assessment of personal or medical history is relatively problematic in elderly men due to age-related changes in performance, and we found the test-retest reliability crucial for our study. One drawback of the test-retest method is that it may be subject to recall bias, since respondents may just repeat what they previously claimed. Socially undesirable items like STDs or sexual behaviors are often under-reported, causing non-differential misclassification. In conclusion, our study uncovered additional protective factors for PCa, including regulating BMI by maintaining good physical activity and reducing total meat consumption. Middle-age and elderly men with PCa family history should stay highly vigilant about the occurrence of PCa. All these results could serve as a basis for designing PCa prevention strategies. It is suggested that clinicians educate patients on up-to-current findings in order to lower risk and promote effective prevention of PCa.

Data availability
All data generated or analysed during this study are included in this published article and its supplementary information files.